The following repository is an implementation of the diffusers library by hugging face and more particularly the pre trained stable diffusion models.

Stable.Diffusion.Demo.mp4

Generated images

Diffusers Library by Hugging Face

The diffusers library by Hugging Face is a powerful tool for implementing diffusion models which are a class of generative models that leverage the concept of diffusion processes to generate high-quality samples. This repo contains the implementation of the Text to Image and Image to Image tasks through stable diffusion models.

Note: For all the pretrained models, I have set the value of safety_checker to None. The documentation does not advice to this, however, I was getting an error Potential NSFW content was detected in one or more images. A black image will be returned instead. Try again with a different prompt and/or seed. and changed the variable as per this issue

Installation

1. Clone this repository

git clone https://github.com/Dhruv16S/Generative-Tasks-through-Stable-Diffusion.git

2. Install required dependencies through

pip install -r requirements.txt

3. Create a file called auth_token.py and define a variable token=<INSERT HUGGING FACE TOKEN HERE>. Specify the user access token in <>. To create a user access token follow the instructions here. Ensure you have a hugging face account before generating your token.

4. Run the application

python app.py

Debugging

If "AssertionError: Torch not compiled with CUDA enabled is encountered, follow the below steps:

1. Run the following command in the same directory

python check_config.py

check the torch version and if CUDA is enabled

2. Make sure you have a compatible NVIDIA GPU and have installed the corresponding CUDA toolkit. You can find the CUDA toolkit version compatible with your GPU on the NVIDIA website

3. Run the following command

pip uninstall torch torchvision -y

4. Install PyTorch with CUDA support by following the instructions (here)[https://pytorch.org/get-started/locally/]. The following command worked for me:

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

5. Re run the flask application

python app.py